Many times I have listened to managers bemoan the fact that the forecasts they are forced to use are not accurate. An accurate forecast is an oxymoron. If a forecast is a prediction of the future, is it realistic to expect it to be accurate? Nonetheless, we have to make forecasts to help prepare our businesses for events that we can not respond to instantaneously. This article is intended to be a basic primer on forecasting as it affects the manufacturing environment. For further information I would refer you to Master Scheduling by John F. Proud and Manufacturing Planning and Control Systems by Vollman, Berry and Whybark. Both of these books can be found in the APICS educational resource listing.
Even if we recognize that forecasts are going to be either a lucky guess or a lousy guess, businesses have to resort to them to be better prepared for future events. Because materials and machine capacity are not instantaneous or infinite, we have to look into the future and guess what events will occur. From these guesses we have to make advance preparations to provide products on time to our customers. The most frequent forecasts are those made of our future sales. From these forecasts, material needs are calculated and purchase orders placed to have materials on hand when the sales orders come in. Sales forecasts can also be the driving force behind plant expansion or contraction. In addition to sales forecasts, there may be forecasts of machine maintenance or break downs; or to help with decisions such as building new plants or adding new equipment. Forecasts can be the prime input to a master schedule.
Forecasting can be a daunting task, but we can make it easier if we look at some basic principles. First, forecasts get less accurate the further out in time we attempt to project. This is obvious because as we look further into the future, there are more unexpected surprises to consider.
Second, forecasts are more accurate if we forecast large groups rather than small groups or individual products. It is intrinsically more accurate to predict the total number of automobiles to be sold in a future period than to predict the number of a specific model. In most manufacturing forecasts the time horizon is relatively short. That is to say less than three years. Projecting into a three year future by product groups will give us a better projection.
Third, forecasts can be generated with computer models. Computers are excellent at managing large amounts of data and using complex algorithms. However, many situations are too complex to be left to a forecast model without some sort of human review. Few companies can afford a computer model that takes into account all the variables in a forecast. Many companies use just a few critical variables and rely on human intelligence to modify the forecast results.
The data base used in a forecast is important. Forecasts can be based on internal data such as the past company history or past sales in a company. They can also be based on external data such as the National Gross Domestic Product (GDP); the number of people in a certain age class; the number of people in a specific income class; or in a certain occupation.
Once we have selected the data to use in our forecast, we have to ask ourselves how we will process the data to project into the future. We could use something as simple as a moving average of a past period of time. We could be more sophisticated and use an exponential smoothing formula to make our projection. Extrapolating from these calculated averages would give us a projection into the future.
Once we have a method for projecting from past data into the future, we need to monitor the bias in our method. This measure of bias can be as simple as measuring the mean error. The mean error is the difference between the actual demand and the forecast demand in each forecast time period, divided by the number of time periods of data. The positive errors are offset by the negative errors. Dividing by the number of time periods yields an average error. The closer the mean error is to zero, the less bias we have in our method.
To determine the size or magnitude of the error is the next measure. An easy to use measure of the size of the error magnitude is the Mean Absolute Deviation (MAD). The MAD defines the size of the error regardless of whether it is positive or negative. The mean error and MAD measures are important because they identify bias in our forecasting method, and give us an idea of the size of the bias.
When forecasting, it is important to manage the forecasting process. You should conduct and review the forecast at regular intervals. Each time you make a forecast; you should be analyzing the result and asking yourself if the results make sense. Use a tracking signal to monitor the quality of the forecast. If the tracking signal is over its boundaries, you have to seriously review the forecast method and results.
In summary, forecasting is an important tool in the business world. It is an especially useful tool in the manufacturing environment, and is often the principal input to a master schedule. Use statistical tools to measure the bias and error in your forecast method. Forecast models are best used to supplement, not replace, management judgment. Apply human input to strengthen the value of the forecast.
You must log in to post a comment.